Two production-oriented Agent Skills for taking a real product from an incomplete idea to a verified Agent product with an explicit, module-by-module implementation strategy.
Official access: sign in, verify eligibility, and download the current packages from CodeSoul Skill Garden.
This public repository is the usage guide and release doorway. It intentionally does not contain the private Skill packages, mirror their source, or provide a direct-download bypass.
| Phase | Skill | Input | Verified output |
|---|---|---|---|
| 1. Define | agent-native-requirements |
An idea, PRD, existing system, or project context | A decision-complete engineering handoff that a development Agent can implement without returning to the user |
| 2. Build | hypha-domain-builder |
The approved handoff, or equivalent complete contracts | A complete Agent product using the approved module routes, with real tests, repair evidence, environment setup, and startup/operations documentation |
The boundary is deliberate. The requirements Skill prepares documents only; the builder consumes approved documents and implements the product. Use phase 1 alone when you only need a specification, or both phases whenever you want the complete product implemented. Hypha is the recommended base, not a requirement.
- Open CodeSoul Skill Garden.
- Sign in with GitHub and complete the displayed access checks.
- Download one or both ZIP packages.
- Extract the package locally. Do not run a Skill directly from inside the ZIP.
- Ask the coding Agent working on your real repository to install the extracted Skill.
- Paste the relevant project prompt below and replace the bracketed text.
The website is the only official download entry. Return there for current releases rather than keeping a hard-coded archive URL.
Help me install the agent-native-requirements Skill from the ZIP I downloaded, then use it to
define my real project: "[project name / problem]". Ask me direct, scenario-based questions until
the requirements are complete enough for an engineering agent to build without coming back to me.
Use this when the idea, PRD, workflow, data contract, risk, quality target, or delivery boundary is not yet fully specified.
Help me install the hypha-domain-builder Skill from the ZIP I downloaded. Read the approved
requirements and module adoption plan for "[project name]", then build, test, fix, and document
the complete Agent product. Use pinned Hypha packages, source, or a hybrid only where the plan
selects them and availability is verified; otherwise follow the approved other-framework or
project-owned route.
Use this only when the engineering handoff is ready, or when you already have an equivalent, approved PRD plus architecture, interface, data, operations, security, and test contracts.
Run these commands from the extracted package root:
./scripts/validate.sh
./scripts/install.sh universal --scope userTo keep the Skill inside one project:
./scripts/install.sh universal --scope project --project-dir /absolute/path/to/projectRestart the Agent or begin a new session after installation so it reloads the Skill inventory. The package README also documents native targets and platform-specific locations for Codex, Claude Code, Kimi Code, OpenCode, Reasonix, OpenClaw, Hermes, Grok Build, Gemini CLI, GitHub Copilot CLI, Trae, WorkBuddy, Qoder, Cursor, Devin, and other compatible Agents.
If your Agent does not support a native Skill directory, ask it to read the extracted SKILL.md
and its referenced files before working. Keep the complete package together; copying only
SKILL.md can remove required templates, references, and validation scripts.
The interview is deliberately direct and scenario-based. It closes the information engineering teams commonly have to rediscover:
- real users, situations, end-to-end journeys, and success evidence;
- application surface: API, CLI, web, mobile, desktop, chatbot, scheduled job, or execution flow;
- exact user/system inputs, outputs, formats, consumers, and failure behavior;
- data sources, databases, data assets, creative assets, permissions, freshness, and ownership;
- target delivery grade and promotion path: prototype, controlled pilot, production, or commercial;
- concurrency, latency, availability, capacity, cost, and stability targets;
- Agent behavior, tools, memory, context, execution, prompt, cache, human approval, and evaluation;
- a module adoption plan covering verified official packages, pinned source, another framework, or intentional project-owned implementation, including upgrade and fallback ownership;
- security, privacy, compliance, observability, operations, migration, rollback, and acceptance;
- unknowns and assumptions that must be closed before implementation.
Choosing a lower grade changes the target rigor, but it does not allow development-critical questions to be skipped.
hypha-domain-builder is an Agent system builder, not a framework generator or test scaffold.
Hypha is the recommended base, but the approved handoff may
select exact official Hypha packages, pinned Hypha source, a controlled hybrid, another framework,
or a frameworkless project-owned implementation for the product and for individual modules.
Official package availability is verified at implementation time; a planned package name is not treated as a published release. The coding Agent clones or updates Hypha only when a source route is selected or a package release must be audited against its source. Every selected route records its version or commit, ownership boundary, tests, upgrade path, fallback, and rollback.
Depending on the approved requirements, it implements and verifies:
- business workflows, runtime/state, events, replay, and recovery;
- governed tools, MCP integrations, execution, memory/context, prompts/models, caches, and policy;
- product-specific API, CLI, web, app, chatbot, or execution surfaces;
- data and asset adapters, environment contracts, secrets handling, and migrations;
- module ownership, dependency provenance, compatibility, upgrade, fallback, and rollback evidence;
- real provider/integration/end-to-end tests, repair and regression loops;
- clean-start verification, release evidence, startup guide, user guide, operations guide, and troubleshooting instructions.
Mocks may support unit tests, but they cannot be used as proof that real integrations or the full user journey work.
- Only documentation is needed: use
agent-native-requirements. - A complete Agent product is needed: use both Skills in order; the approved module plan decides whether it uses Hypha packages, Hypha source, a hybrid, another framework, or no framework.
- Approved equivalent documents already exist: start with
hypha-domain-builder; it will block rather than invent a material missing contract. - The product will not use Hypha: the same builder remains applicable and must follow the approved other-framework or project-owned routes.
- Download the Skills
- Hypha — Agent-native runtime and framework
- Plasmod — Agent-native data and evidence substrate
- CodeSoul on Hugging Face
- CodeSoul on GitHub
Found an unclear instruction, an installation issue, a useful real-world scenario, or an idea that would make either Skill more helpful? We would like to hear it.
- Email CodeSoul at openex@code-soul.com.
- Use this public repository's Issues for public documentation feedback and reproducible usage reports.
- Read CONTRIBUTING.md before proposing a documentation change.
Please never include passwords, API keys, private project data, customer information, or private Skill package contents in an issue, pull request, or email.
Developed and maintained by CodeSoul.
This repository contains public documentation only. The downloadable Skill packages are source-available software distributed under the PolyForm Strict License 1.0.0 and are limited to permitted noncommercial use. The license included in each downloaded package is authoritative. Public visibility of this guide does not grant permission to copy, redistribute, modify, or commercially use the private Skill packages.